III. Proposed Algorithm
3.5 Final Enhanced Image
From equation (3.13), we obtain the enhanced illumination 𝐿̃𝑙𝑢𝑚. We already obtained the reflectance 𝑅𝑙𝑢𝑚 using ABF in Section 3.2 and equation (3.6). Therefore, the enhanced Y channel 𝐼𝑒𝑛ℎ is obtained by 𝐿̃𝑙𝑢𝑚 and 𝑅𝑙𝑢𝑚 using equation (3.1), as follows:
𝐼𝑒𝑛ℎ(𝑥, 𝑦) = 𝑅𝑙𝑢𝑚(𝑥, 𝑦) ∙ 𝐿̃𝑙𝑢𝑚(𝑥, 𝑦). (3.14) Finally, the image is obtained by converting the enhanced 𝐼𝑒𝑛ℎ and original u and v to the enhanced RGB color space.
20
Chapter IV
Experimental Results
Our experiment used many low-contrast images to evaluate changes in probability distributions and cumulative distributions. Our algorithm was also compared to others, including those by Shen et al. [8]
and Shuhang et al. [12]. All parameters used in our algorithm were determined experimentally and were fixed, such as 𝜏𝑐ℎ𝑟 = 0.025 and 𝜏𝑐 = 0.006. 𝜎1 and 𝜎2 had standard deviations of 3 and 5, respectively. Exception standards for estimating the low-contrast regions were also determined experimentally. The size of the adaptive bilateral filter window was 11× 11 for smoothing the neighbor pixels.
We applied A-law based tone mapping without and with weight in seven input images. The results of this application are shown in Figures 4.1–4.7. The low-contrast region of Figures 4.1(a), 4.2(a), 4.3(a), and 4.4(a) were placed in the low-intensity. Meanwhile, the low-contrast region of Figure 4.5(a), and 4.6(a) were placed in the middle-intensity and the low-contrast region of Figure 4.7(a) was placed in the high-intensity. Figures 4.1(c), 4.2(c), 4.3(c), and 4.4(c) show conventional A-law based tone mapping results. Figures 4.1(d), 4.2(d), 4.3(d), and 4.4(d) show weight A-law based tone mapping results. From these results, it can be concluded that weight A-law based tone mapping creates greater enhancement in images than conventional A-law based tone mapping in terms of visibility. Figures 4.1(e) and (f), 4.2(e) and (f), 4.3(e) and (f), and 4.4(e) and (f) show that weight A-law based tone mapping is more enhancive than conventional A-law based tone mapping in terms of probability distribution and cumulative distribution. Figures 4.5(c), (d), (e), and (f); 4.6(c), (d), (e), and (f); and 4.7(c), (d), (e), and (f) show that conventional A-Law tone-mapping is better than weight A-law based tone mapping in terms of visibility, but that weight A-law based tone mapping is more enhancive than conventional A-law based tone mapping in terms of probability distribution and cumulative distribution.
From the experimental results, we can see the importance of the low-contrast region's location. If the low-contrast region is placed in a low-intensity region, the use of weighted A-law based tone mapping is most appropriate, especially in terms of visibility, probability distribution, and cumulative distribution. If the low-contrast region is placed in a middle or high intensity region, then the use of A- law based tone mapping is more appropriate in terms of visibility, but the weighted A-law based tone mapping is better in terms of the other factors.
The results of comparison with this application and other algorithms are shown in Figures 4.8–4.14.
The method used by Shen et al. [8] is not ideal for enhancing all types of low-contrast image. As
21
shown in Figures 4.8(c), 4.9(c), 4.10(c), and 4.11(c), Shen et al.'s method [8] enhances low-intensity images; however, as shown in Figures 4.12(c), 4.13(c), and 4.14(c), this method is not appropriate for use with middle or high intensity, low-contrast image enhancement. Shen et al.'s algorithm [8] increases intensity to enhance low-contrast images, and this algorithm uses fixed tone-mapping despite the fact that the degree of darkness is different in every image. Thus, this method does not enhance all kinds of low-contrast images.
The method proposed by Shuhang et al. [12] efficiently enhances all types of low-contrast images while still preserving naturalness. Shuhang et al.'s algorithm [12] uses cumulative distribution matching to solve contrast problems. However, similar illumination values in the same low-contrast region can be changed too much by cumulative distribution matching. This causes distortion in the enhanced image.
As shown in Figures 4.10(d), 4.11(d), 4.12(d), 4.13(d), and 4.14(d), Shuhang et al.'s method [12]
preserves the original tone of the image and enhances it as well, but Figures 4.8(d) and 4.9(d) show that a distortion problem has occurred.
22
(a) (b)
(c) (d)
(e) (f)
Figure 4.1: Experimental results of the proposed algorithm on the House image. (a) An input low contrast image, (b) contrast region labeling, the enhanced images (c) without and (d) with the weighting scheme for A-law based tone mapping, and (e) the changed probability distributions and the (f) changed cumulative distributions without (up) and with (below) the weighting scheme.
23
(a) (b)
(c) (d)
(e) (f)
Figure 4.2: Experimental results of the proposed algorithm on the Dog image. (a) An input low contrast image, (b) contrast region labeling, the enhanced images (c) without and (d) with the weighting scheme for A-law based tone mapping, and (e) the changed probability distributions and the (f) changed cumulative distributions without (up) and with (below) the weighting scheme.
24
(a) (b)
(c) (d)
(e) (f)
Figure 4.3: Experimental results of the proposed algorithm on the Village image. (a) An input low contrast image, (b) contrast region labeling, the enhanced images (c) without and (d) with the weighting scheme for A-law based tone mapping, and (e) the changed probability distributions and the (f) changed cumulative distributions without (up) and with (below) the weighting scheme.
25
(a) (b)
(c) (d)
(f) (g)
Figure 4.4: Experimental results of the proposed algorithm on the Baby image. (a) An input low contrast image, (b) contrast region labeling, the enhanced images (c) without and (d) with the weighting scheme for A-law based tone mapping, and (e) the changed probability distributions and the (f) changed cumulative distributions without (up) and with (below) the weighting scheme.
26
(a) (b)
(c) (d)
(e) (f)
Figure 4.5: Experimental results of the proposed algorithm on the Diver image. (a) An input low contrast image, (b) contrast region labeling, the enhanced images (c) without and (d) with the weighting scheme for A-law based tone mapping, and (e) the changed probability distributions and the (f) changed cumulative distributions without (up) and with (below) the weighting scheme.
27
(a) (b)
(c) (d)
(e) (f)
Figure 4.6: Experimental results of the proposed algorithm on the Foggy image. (a) An input low contrast image, (b) contrast region labeling, the enhanced images (c) without and (d) with the weighting scheme for A-law based tone mapping, and (e) the changed probability distributions and the (f) changed cumulative distributions without (up) and with (below) the weighting scheme.
28
(a) (b)
(c) (d)
(e) (f)
Figure 4.7: Experimental results of the proposed algorithm on the Airplane image. (a) An input low contrast image, (b) contrast region labeling, the enhanced images (c) without and (d) with the weighting scheme for A-law based tone mapping, and (e) the changed probability distributions and the (f) changed cumulative distributions without (up) and with (below) the weighting scheme.
29
(a) (b)
(c) (d)
Figure 4.8: Comparison of enhancement results on the House image. (a) An input low contrast image.
The enhanced images by using (b) the proposed algorithm, (c) Shen et al.’s algorithm, and (d) Shuhang et al.’s algorithm, respectively.
(a) (b) (c) (d)
Figure 4.9: Comparison of enhancement results on the Dog image. (a) An input low contrast image.
The enhanced images by using (b) the proposed algorithm, (c) Shen et al.’s algorithm, and (d) Shuhang et al.’s algorithm, respectively.
30
(a) (b)
(c) (d)
Figure 4.10: Comparison of enhancement results on the Village image. (a) An input low contrast image. The enhanced images by using (b) the proposed algorithm, (c) Shen et al.’s algorithm, and (d) Shuhang et al.’s algorithm, respectively.
(a) (b)
(c) (d)
Figure 4.11: Comparison of enhancement results on the Baby image. (a) An input low contrast image.
The enhanced images by using (b) the proposed algorithm, (c) Shen et al.’s algorithm, and (d) Shuhang et al.’s algorithm, respectively.
31
(a) (b)
(c) (d)
Figure 4.12: Comparison of enhancement results on the Diver image. (a) An input low contrast image.
The enhanced images by using (b) the proposed algorithm, (c) Shen et al.’s algorithm, and (d) Shuhang et al.’s algorithm, respectively.
(a) (b)
(c) (d)
Figure 4.13: Comparison of enhancement results on the Foggy image. (a) An input low contrast image. The enhanced images by using (b) the proposed algorithm, (c) Shen et al.’s algorithm, and (d) Shuhang et al.’s algorithm, respectively.
32
(a) (b)
(c) (d)
Figure 4.14: Comparison of enhancement results on the Airplane image. (a) An input low contrast image. The enhanced images by using (b) the proposed algorithm, (c) Shen et al.’s algorithm, and (d) Shuhang et al.’s algorithm, respectively.
33
Chapter V Conclusion
In this thesis, we proposed a low-contrast image enhancement algorithm using adaptive bilateral filtering and weighted A-Law tone-mapping. We solved halo artifact problem using adaptive bilateral filtering and preserved naturalness using adaptive A-law based tone mapping. Also, our algorithm solved a low-contrast image enhancement problem in terms of probability and cumulative distribution.
The experimental results showed that the proposed algorithm enhances low contrast images efficiently.